Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
The Definition People Get Wrong
Cannibalization is not two pages mentioning the same phrase. It is two pages competing for the same query in a way that leaves both weaker: references and engagement divided, neither page accumulating enough to compete, and Google choosing between them in a way you did not intend and cannot predict.
The distinction matters because the loose definition generates enormous amounts of unnecessary work. Any site with more than fifty pages will have dozens of overlapping phrases, and tools that flag overlap rather than harm will report all of them. Acting on that list means merging pages that were fine, and consolidation is not free: you lose URLs, references, and sometimes rankings that were working.
It is also worth stating what Google does not do. There is no cannibalization penalty. The system picks a page it considers most relevant for each query and shows that one. The damage is entirely second-order, which is why the diagnosis has to be evidential rather than structural, and why "these two pages are about the same thing" is not a finding.
You can name the specific harm each flagged cluster is causing, or you close it as a false positive.
Finding It With Data You Already Have
One export does the whole job: Search Console performance, Query and Page dimensions together, the longest window available. You are looking for queries whose impressions are distributed across multiple URLs rather than concentrated on one.
- Export query and page together.
Not query alone, which hides the split entirely, and not page alone, which hides which query is contested. The pairing is the only view that shows the problem, and it is why tool alerts built on rank tracking miss most real cases.
- Group by query and count distinct URLs.
A pivot table is sufficient. Keep only queries where two or more URLs received impressions, and where the total impressions are large enough to matter to you. A contested query with eleven impressions is not a project.
- Check whether the split is meaningful.
A ninety-five to five split is Google being confident, and it is fine. A sixty-forty split is a genuine contest. The middle is where the harm concentrates, because neither page is accumulating the signals that would settle it.
- Look at position, not just impressions.
Two pages splitting a query while one sits at position four is not urgent. Two pages splitting a query while both hover between eleven and twenty is the classic pattern: divided effort, neither crossing the threshold into real traffic.
- Check stability over time.
Compare two adjacent periods. If the URL receiving impressions for a query changed between them, Google is undecided, and an undecided selection is the strongest single signal that intervention will help.
Telling Real Cases From False Positives
Ask one question of every flagged cluster: do these pages serve different intents behind the same words? If yes, the overlap is legitimate and you should leave it alone. If they serve the same intent with different words, you have found something worth fixing.
| Pattern | Verdict | Why |
|---|---|---|
| Category page and buying guide share a commercial query | Leave alone | Different moments in one decision; both can legitimately appear |
| Three blog posts answering the same question | Real cannibalization | Same intent, split signals, none authoritative |
| Service page and location page for the same service | Usually leave alone | Geographic intent differs; see multi-location SEO |
| Old post and its updated replacement both live | Real, and easy | One was meant to supersede the other and never did |
| Paginated pages sharing the parent query | Not cannibalization | A pagination handling question, covered in the pagination guide |
| Tag or archive pages competing with articles | Real, and structural | Thin index pages should usually not be indexable at all |
That last row is the one worth checking first on any content-managed site, because it is systemic rather than per-page. Automatically generated tag, author, and date archives frequently compete with the articles they list, and the fix is a template decision made once rather than a cluster-by-cluster negotiation.
The Four Fixes, in Order of Reversibility
Always prefer the reversible fix. Internal links can be changed back in an afternoon; a redirect and a deleted page cannot. Work down this list and stop at the first option that resolves the contest.
- Repoint internal links.
Make every internal link about that topic point at the page you want to win, with anchor text that describes it. This is frequently sufficient on its own, because inconsistent internal linking is what created the ambiguity in the first place.
- Differentiate the pages.
If both pages should exist, make them visibly answer different questions: change the titles, the opening paragraph, and the headings so the distinction is legible to a reader in three seconds. If you cannot articulate the difference, that is the answer to whether both should exist.
- Merge, then redirect.
Move the unique substance from the losing page into the winner first, then redirect. Redirecting before merging throws away the only thing that made the loser worth keeping, and this is the most common way consolidation loses traffic.
- Noindex, for pages with a non-search purpose.
A page that exists for customers but not for search, such as a thin archive or a duplicate print view, can be removed from the index without being removed from the site. The distinction between this and a robots.txt block matters and is set out in robots.txt vs noindex vs canonical.
Whichever you choose, change one cluster at a time and record the date. Cannibalization fixes take weeks to resolve, and if you change twenty clusters in one afternoon you will never know which intervention worked. The consolidation mechanics in detail, including how to preserve references, are in the content consolidation playbook.
Why Internal Links Come Before Redirects
Your internal linking is the clearest statement you make about which page matters for a topic. When forty links point at one page and thirty at another with similar anchor text, you have told the search engine you are undecided, and it responds by being undecided too.
This is why the link fix so often resolves the contest without any URL changes. It is also why the problem recurs on sites that publish steadily without a linking convention: every new post links to whichever older post the author remembered, and the distribution drifts. A convention as simple as "one canonical page per topic, and everything on that topic links to it" prevents most of this permanently.
Do the link work with a map rather than by memory, because the pages you forget are the ones causing the split. The method is in the internal linking strategy guide, and the internal link planner will show you where the current links actually point rather than where you assume they do.
Preventing the Next One
Cannibalization is a planning failure that shows up as a technical symptom. It happens when nobody checks whether a page already exists before commissioning a new one, which is a five-minute check and almost never done.
Two habits remove most future cases. Before writing anything, search your own site for the target query and read what comes back; if a page already covers it, the correct action is usually to improve that page rather than add another. And keep one list of which page owns which topic, updated when you publish, so the question "where does this belong" has an answer that does not depend on who is asking.
The related discipline is deciding what deserves to exist at all, which is upstream of both cannibalization and the tail-page problem described in page-level authority. Sites that publish against a topic map rather than a calendar rarely develop this problem, and the planning frame for that is in topical authority.
Questions People Ask About Keyword Cannibalization
- What is keyword cannibalization?
It is when two or more pages on your site compete for the same query in a way that leaves both worse off: ranking signals split between them, the wrong page surfaces, or the URL Google shows keeps flipping. It is not simply two pages containing the same phrase, which is normal and harmless on any site of reasonable size.
- How do I find keyword cannibalization in Search Console?
Export performance data with both Query and Page dimensions over a long window, then look for queries where impressions are split across two or more URLs. The signature worth acting on is a single query sending meaningful impressions to multiple pages, especially where the URL receiving them changes over time and neither page reaches a strong position.
- Does keyword cannibalization actually hurt rankings?
Sometimes, and less often than it is diagnosed. Google generally picks one page per query and does not apply a penalty for having several on a topic. The genuine harm is indirect: your links, references, and engagement are divided across pages, so none accumulates enough to compete, and Google may select a page you would not have chosen.
- Should I delete or redirect the losing page?
Usually neither, at least not first. Fixing internal links so they point consistently at the page you want to win often resolves the signal split without touching URLs. Redirect only when the loser has no independent purpose, and merge its unique content into the winner before you do, so the value moves rather than disappearing.
- Can two of my pages legitimately rank for the same keyword?
Yes, and it is often desirable. A category page and a buying guide can both appear for the same commercial phrase because they serve different moments in the same decision. The test is not whether two pages share a query but whether they serve different intents behind it. If they do, leave them alone.
